A nutrition-based radiomics–clinical model to predict the prognosis of patients with acute-on-chronic liver failure

• Published in: Displays Vol. 84; p. 102750
• Main Authors: Zhang, Qian, Peng, Yunsong, Lei, Siyi, Xiong, Tingting, Zhang, Liulu, Peng, Hong, Luo, Xinhua, Wang, Rongpin
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2024
• Subjects: End-stage liver disease; LASSO regression; Predictive model; Radiomics; Skeletal muscles; LASSO regression; Skeletal muscles; End-stage liver disease; Predictive model; Radiomics
• ISSN: 0141-9382
• DOI: 10.1016/j.displa.2024.102750

•It's the first time a study has extracted radiomics features from muscle and fat to reflect the patient’s nutritional status.•We combined nutrition-based radiomics features with clinical markers to predict the 90-day prognosis of patients with ACLF.•The 90-day prognosis of the low- and high-risk groups were well distinguished by this radiomics–clinical model. Acute-on-chronic liver failure (ACLF) lead to high short-term mortality. Effective identification of patients with poor prognoses and early intervention are crucial. Meanwhile, the nutritional status of patients is vital in the prognosis of chronic liver diseases. Hence, the study applies artificial intelligence to extract nutrition-based radiomics features in ACLF patients and construct a radiomics-clinical model to predict the 90-day prognosis. 163 ACLF patients were recruited from our hospital. The patients were randomly assigned to the training cohort (n = 99) and the test cohort (n = 64). The skeletal muscle, subcutaneous fat and visceral fat radiomics features were extracted. The blood markers were collected. The LASSO regression model was applied to select radiomics features and form the best feature subset. Finally, the linear discriminant classifier was used for endpoint prediction. In both the training and test cohorts, 30.3 % and 29.7 % of patients, respectively, experienced disease progression. In the training cohort, this progression included 23.2 % experiencing death and 7.1 % undergoing liver transplantation. Similarly, in the test cohort, 20.3 % experienced death, while 9.4 % underwent liver transplantation. LASSO regression screened 15 features for constructing a radiomics-clinical model, 6 of 15 features were texture features of skeletal muscles. The area under the curve (AUC) of this model in the test cohort was 0.873 (95 % CI 0.773–0.972). To use Delong test compared the AUC of the radiomics-clinical model with sub-models (radiomics model and clinical model) and traditional classical models (MELD and MELD-Na). The results showed that the radiomics-clinical model was superior to any of the above models (p < 0.05). The nutrition-based radiomics-clinical machine learning model can effectively predict the 90-day adverse prognosis of ACLF patients.